Chapter 2

The Hodgkin–Huxley Theory of Neuronal

Excitation

Hodgkin and Huxley (1952) proposed the famous Hodgkin–Huxley (hereinafter

referred to as HH) equations which quantitatively describe the generation of action
potential of squid giant axon, although there are still arguments against it (Connor
et al. 1977; Strassberg and DeFelice 1993; Rush and Rinzel 1995; Clay 1998). The
HH equations are important not only in that it is one of the most successful math-
ematical model in quantitatively describing biological phenomena but also in that
the method (the HH formalism or the HH theory) used in deriving the model of a
squid is directly applicable to many kinds of neurons and other excitable cells. The
equations derived following this HH formalism are called the HH-type equations.
The dynamical system theory is very useful to analyze and understand the
dynamics of the HH equations. On the other hand, the HH equations have rich math-
ematical structures and give many insights to mathematics also. For example, there
are still many studies and mathematical findings on the “classical” HH equations
(Plant 1976; Rinzel 1978; Hassard 1978; Troy 1978; Rinzel and Miller 1980;
Labouriau 1985, 1989; Hassard and Shiau 1989; Shiau and Hassard 1991; Bedrov
et al. 1992; Guckenheimer and Labouriau 1993; Labouriau and Ruas 1996; Fukai
et al. 2000a,b). This chapter gives an overview of the dynamics of the HH equations
and of the mechanism of the action potential generation.

2.1 What is a Neuron? Neuron is a Signal Converter

Figure 2.1 illustrates a shape and a function of neurons schematically. Our brain is
a complicated network of a tremendous number of neurons. Right figure shows a
small network of three neurons. A neuron has a very special shape which is much
different from usual sphere-shaped or disklike cells. A soma is the main body of
neuron from which a long cable called an axon is extended. Neurons transmit and
exchange electric signals called action potentials or spikes, each other. (The gener-
ation of a spike is also called as the excitation or the firing of a neuron.) Neurons
receive the spikes at a synapse which is a connection between neurons. Then, the
electric signals or information is transmitted in the direction from a dendrite to an
axon. The upper-left panel of Fig. 2.1 illustrates the waveform of action potentials.
Action potential or a spike has an amplitude of about 100 mV.

S. Doi et al., Computational Electrophysiology, 37

DOI 10.1007/978-4-431-53862-2 2,  c Springer 2010
38 2 The Hodgkin–Huxley Theory of Neuronal Excitation

v (mV)
100 Neurons
80
Action potential
60
40
20
0
–20 Inputs to a neuron
–40 Axon
0 10 20 30 40 terminal
time (ms) Axon

Soma Synapse
Dendrite
?
Neuron

Fig. 2.1 Diagrams illustrating: A network of three neurons which exchange electric signals called
action potentials, each other (right). Waveform of action potentials (upper left). Neuron as a device
which converts input signals to output signals (lower left)

Typical neurons do not generate any spikes without input signals (i.e. spikes from
other neurons). A sufficiently large input pulse causes a neuron to generate an output
spike, as illustrated in the upper-left panel, whereas no output spike is generated by
a small input (the first pulse in the lower trace of the upper-left panel). Therefore, a
neuron possesses a threshold or all-or-none characteristic. There is a special period
or timing called the refractory period (the timing of the downstroke of the action po-
tential) in which the neuron cannot produce any output spike even though sufficient
amount of inputs (the third and fourth pulses in the panel) were put in the neuron.
Thus, we can consider a neuron as a device which transforms or converts the train
of input spikes to a train of output spikes (see the lower-left panel of Fig. 2.1).

2.2 The Hodgkin–Huxley Formulation of Excitable

Cell Membranes

This section briefly explains the framework of the HH formalism to model the action
potential generation of neurons and of other excitable cells.
Biological cells, including neurons, are enclosed by a plasma membrane or sim-
ply membrane which separates the intracellular and extracellular water-containing
media. The cell membrane consists of lipid bilayer, as shown in Fig. 2.2. There
are various ions in both the intra- and extra-cellular regions. The concentrations
of ions, however, are much different between the intra- and extra-cellular regions.
For example, the concentration of KC ion is high and low in the intra- and extra-
cellular regions, respectively. On the contrary, that of NaC ion is low and high
2.2 The Hodgkin–Huxley Formulation of Excitable Cell Membranes 39

Ion-selective filter
Membrane potential

K+ chanel
Na+ 460 [mM]
Cell membrane
K+ 10 K+ 400 Gate Cell membrane
(Lipid bilayer)
Cl– 540 Na+ 50 Cl– 70

– 70 [mV] Outside the cell membrane

+
Na channel Leak channel
ENa EK

Capacitance
Concentraton distribution of ions ......
inside and outside the cell
gNa gK

Inside the cell membrane

Fig. 2.2 The equivalent-circuit formulation of a cell membrane and ionic channels by Hodgkin
and Huxley

Na channel K channel Open and close of a gate

dm
= (1 – m) am(V) – mbm(V )
dt
m gates n gates
bm(V )
Open Close
h gate am(V )

Fig. 2.3 Diagrams explaining the gate dynamics

in the intra- and extra-cellular regions, respectively. Usually, the resistance of the
membrane is very high and the membrane acts as an insulator to the movement of
ions. If the electrical potential at the inside surface of the cell membrane is com-
pared to the potential at the outside surface, there is a potential difference or voltage
called the transmembrane potential or simply the membrane potential.
In the membrane, there are holes through which ions can move in and out. Such
a hole is called an ion channel and consists of membrane proteins. Ion channels
are not simple holes (pores) or passive resistors through which ion flux flows. Ion
channels are selective for a particular ion. For example, an ionic channel named NaC
channel can pass only NaC ions. Also, ion channels are dynamic and sensitive to the
membrane potential and to other factors. Namely, they open and close depending
on such factors. An ion channel has several gates and the opening and closing of
the ion channel are controlled by the gates as shown in Fig. 2.3. Note that, there
40 2 The Hodgkin–Huxley Theory of Neuronal Excitation

are various types of ionic channels which pass a specific ion. Particularly, there are
many variants of KC channels classified by their various characteristics (Adams
1982; Crill and Schwindt 1983; Llinas 1988; Hille 1992).
The basic idea of the HH formalism is to just recognize the cell membrane as a
simple electric circuit as shown in the lower-right panel of Fig. 2.2. The capacitive
property of the cell membrane is denoted by the capacitor with a certain capaci-
tance in the circuit. NaC and KC channels are modeled by the resistors which have
conductances gNa and gK , respectively. Note that the resistors are not linear but non-
linear, and also are dynamic: the values of the conductances vary temporally, which
are explained later. There is a tendency that NaC ions flow inward and that KC ions
flow outward the cell membrane because there are differences in the concentrations
of the ions between inside and outside of the membrane. Namely, ions have a ten-
dency to move down their concentration gradients. Such a tendency is denoted by
the batteries with voltages ENa and EK in the circuit. These voltages depend on the
inside–outside concentration difference in each ion. Notice that the polarities of the
batteries ENa and EK are reversed.

2.3 Nonlinear Dynamical Analysis of the Original

HH Equations

The Hodgkin–Huxley equations (Hodgkin and Huxley 1952) of a squid giant axon
are simply the differential equations of the electric circuit shown in Fig. 2.2 and are
described as follows:

@v @2 v
C D a C G.v; m; n; h/ C Iext ; (2.1a)
@t 2 @x 2
@m
D ˛m .v/.1  m/  ˇm .v/m; (2.1b)
@t
@n D ˛ .v/.1  n/  ˇ .v/n; (2.1c)
n n
@t
@h
D ˛h .v/.1  h/  ˇh .v/hI (2.1d)
@t

G.v; m; n; h/ D INa .v; m; h/ C IK .v; n/ C IL .v/

D gN Na m3 h.VNa  v/ C gN K n4 .VK  v/ C gN L .VL  v/I (2.2)

C D 1 F cm2 ; gN Na D 120 mS cm2 ; gN K D 36 mS cm2 ; gN L D 0:3 mS cm2 ;

VNa D 115 mV; VK D 12 mV; VL D 10:599 mVI
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 41

0:1.25  v/
˛m .v/ D ; ˇm .v/ D 4e v=18 ;
expŒ.25  v/=10  1
0:01.10  v/
˛n .v/ D ; ˇn .v/ D 0:125e v=80 ;
expŒ.10  v/=10  1
˛h .v/ D 0:07 expŒv=20; ˇh .v/ D 1 I
expŒ.30  v/=10 C 1
where v (mV) is the membrane potential. Equation (2.1a) simply denotes the
Kirchhoff’s law. (In (2.1), a neuron is considered as a cylinder-shaped cell and
the dependence of v on its position x is also taken into account. The term
.a=2/.@2 v=@x 2 / denotes the diffusions of ions along the axis of the cylinder.
However, either the shape of neurons or the x-dependence are not considered in
this book.) INa and IK are the currents through NaC and KC channels, respectively.
The current IL is the leak current and denotes all residue currents through a cell
membrane other than NaC and KC currents.
As seen from (2.2), the NaC current INa is denoted by gN Na m3 h.VNa  v/ which
takes a form of (Conductance)  (Voltage): Ohm’s law. The voltage VNa is called
the Nernst potential or the equilibrium potential or sometimes the resting potential
(do not confuse with the resting potential of whole membrane) of NaC ion. The
Nernst potential is the potential where the tendency of ions to move down their
concentration gradient is exactly balanced with the force by the electric potential
difference; no NaC current flow through the NaC channel when v D VNa . gN Na m3 h
(D gNa in the circuit of Fig. 2.2) denotes the conductance of NaC channel where the
constant gN Na is called the maximum conductance of the channel and m3 h denotes
dynamic or temporal change of the conductance. In the KC current IK , the term n4
denotes the temporal change of KC channel conductance.
The variables m, n and h take a (dimensionless) value between zero and unity,
and are called the gate variables. As seen from the left panel of Fig. 2.3, it is as-
sumed that NaC channel possesses three m-gates and single h-gate whereas KC
channel four n-gates. In the HH formalism, it is also assumed that the variables m,
n and h denote the probabilities that corresponding gates are open. The dynamic
opening and closing of gates obey a simple (linear) process described by (2.1b–d)
where ˛m .v/ (ˇm .v/) is the rate “constant” for changing from closed (opened) state
to opened (closed, resp.) state, as shown in the right panel of Fig. 2.3. Note that the
rate “constant” ˛m .v/ and ˇm .v/ are not actually the constants but the functions
which depend on the membrane potential v. m and h are also called the activation
and inactivation variables of NaC ionic channel, respectively, while n the activation
variable of KC channel. The reason of this naming is because m and n are the in-
creasing functions of v while h the decreasing one. Iext ( A cm2 ) is the constant
current externally applied to a neuron. The constants a and  are the radius and
resistivity of the “cylindrical” axon, respectively. Throughout this book, we do not
treat either such a cylindrical axon or the partial differential equation (2.1) which is
an infinite-dimensional dynamical system. For the rich and complicated dynamics of
such neuronal cable equations, see Carpenter (1977), Horikawa (1994), Kepler and
Marder (1993), Rinzel and Keener (1983), Poznanski (1998), and Yanagida (1985,
1987, 1989).
42 2 The Hodgkin–Huxley Theory of Neuronal Excitation

2.3.1 Simple Dynamics of Gating Variables

In the following, we assume that the membrane potential v is spatially constant and
omit the spatial derivative in (2.1); we consider the following HH equations for a
space-clamped squid giant axon:

dv
C D G.v; m; n; h/ C Iext ; (2.3a)
dt
dm 1
D .m1 .v/  m/; (2.3b)
dt
m .v/
dn
D 1 .n1 .v/  n/; (2.3c)
dt
n .v/
dh
D 1 .h1 .v/  h/; (2.3d)
dt
h .v/

where

1 ˛x .v/

x .v/ D ; x 1 .v/ D ; x D m; n; h: (2.4)
˛x .v/ C ˇx .v/ ˛x .v/ C ˇx .v/

(The word “space-clamped” means that both the shape of a neuron and the depen-
dence of the membrane potential v on the spacial position x are ignored while both
were taken into account in (2.1).)
Figure 2.4 shows an example of a numerically solved solution of the HH
equations (2.3). Panel (a) is a waveform of the membrane potential. A pulsatile
input applied at a time t D 5 ms induces an action potential. Panel (b) is the wave-
forms of the gating variables m, n and h. Total membrane current, Na current, K
current and leak current are shown in (c)–(f), respectively.
The HH equations (2.3) are nonlinear differential equations with four variables
and apparently look very complicated. Equations (2.3b–d) which describe the dy-
namics of gating variables, however, share a simple common structure. Functions

x .v/ and x 1 .v/, x D m; n; h depend on the membrane potential v and thus vary
temporally with the temporal change of v. If we assume that the functions do not de-
pend on v (
x .v/ 
x , x 1 .v/  x 1 ), then (2.3b–d) reduce to a linear differential
equation
dx 1
D
.x 1  x/; x D m; n; h;
dt x

whose analytical solution with an initial value x D x0 is

x.t/ D exp.t=
x /.x0  x 1 / C x 1 :
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 43

a b
v (mV) Iext (µA/cm2)
100 50 1.0
80 40 0.8
60
30 0.6
40
20 20 0.4
0 10 0.2
–20
0
0 5 10 15 20 25 0 5 10 15 20 25
c 2
d
G (µA/cm ) INa (µA/cm2)
250
200 600
150
100 400
50
0 200
–50
0
0 5 10 15 20 25 0 5 10 15 20 25
e f
2 2
IK (µA/cm ) IL (µA/cm )
0 5
0
–200
–5
–400 –10
–15
–600 –20
–25
–800
0 5 10 15 20 25 0 5 10 15 20 25
t (ms) t (ms)

Fig. 2.4 Solution of the Hodgkin–Huxley equations

The variable x.t/ approaches x 1 at a speed depending on a time constant

x .
In (2.3b–d), although they are function of v and we cannot solve the equation
analytically,
x .v/ still preserves a role of “time constant” and x 1 .v/ is the
steady-state function to which the variable x asymptotically approaches in a steady
(or stationary) state.
Figure 2.5a shows the functions x 1 .v/; x D m; n; h. In spite of the complicated
functional forms, these functions are all monotonic functions and have the shape of
so-called “sigmoid function.” m1 .v/ and n1 .v/ are increasing functions and thus
variables m and n are activation variables while h1 .v/ is a decreasing function and
h is an inactivation variable. Figure 2.5b shows the functions
x .v/; x D m; n; h
which change depending on v. The function
m .v/, however, much smaller than

n .v/;
h .v/ in the whole range of v. In the following, let us explore the dynamics of
the HH equations regarding this time-scale difference.
44 2 The Hodgkin–Huxley Theory of Neuronal Excitation

a b
1.0
n
0.8 h 8 τh

0.6 6
m τn
0.4 4
τm
0.2 2
0.0 0
–100 –50 0 50 100 –50 0 50 100 150
v (mV) v (mV)

Fig. 2.5 Various functions in the Hodgkin–Huxley equations

2.3.2 FitzHugh’s Subsystem Analysis of the HH Equations

In this subsection, following the pioneering paper FitzHugh (1960), let us see from
what dynamics the threshold property of a neuron or the HH equations comes.
The HH equations have four variables and it is difficult to observe the full (four-
dimensional) state space directly. The separation of the full state space to several
subspaces, however, resolves this difficulty. As shown in Fig. 2.5, the “times con-
stants” of variables n and h are much bigger than that of m; n and h change their
values more slowly than m. Thus we (temporarily) ignore the dynamics of n and h
in the HH equations and consider the following system:

dv
C D gN Na m3 h.VNa  v/ C gN K n4 .VK  v/ C gN L .VL  v/; (2.5a)
dt
dm
D 1 .m1 .v/  m/: (2.5b)
dt
m .v/

(We call this system as the v–m subsystem.) In the v–m subsystem, v and m are the
dynamic variables while h and n are set in suitable values as a “parameter.”
By using the v–m subsystem, let us explain the firing process in the HH equations
in the following order:

quiescent state ! depolarization ! decrease of h ! increase of n

! repolarization :

Figure 2.6a shows the v–m phase plane of (2.5) when the values of h and n are fixed
to that of the quiescent state (a stable equilibrium point) of the HH equations (2.3).
The m-nullcline (a curve in which d m=dt D 0: m D m1 .v/) is a sigmoidal mono-
tonically increasing function. (For more explanation of nullclines, see Chap. 3.) We
can see that the v-nullcline (dotted curve) and m-nullcline (solid curve) intersect in
the three points v1 , v2 and v3 which are equilibrium points of the v–m subsystem
(panel b is the magnification of lower-left region of a). The leftmost equilibrium
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 45

a m b m
1.0 100
v∗
dm =0
3 v∗2 dv
=0
0.8 80
dt dt
0.6 60

×10–3
v∗1
0.4 40
dm
0.2 v∗2 dv 20 =0
v∗1 =0 dt
dt
0.0 0
0 20 40 60 80 100 0 2 4 6 8 10
v (mV) v (mV)

c m
d m
1.0 dm 1.0
=0 v∗
dt 3 dm = 0
0.8 0.8 dt
dv =0
0.6 0.6 dt
dv
v∗2 =0
dt
0.4 0.4

0.2 ∗ 0.2 v∗1

v1
0.0 0.0
0 20 40 60 80 100 0 20 40 60 80 100
v (mV) v (mV)

Fig. 2.6 Phase plane of the v–m subsystem (2.5) of the HH equations. (a) n D 0:317677,
h D 0:596120. (b) Magnification of (a). (c) n D 0:317677, h D 0:02. (d) n D 0:5, h D 0:02

point v1 corresponds to the quiescent state (a stable equilibrium point) and the mid-
dle point v2 is a saddle point whose stable manifold (the broken curve tending to
v2 ) forms a threshold between exciting and non-exciting, which means that the HH
model when h and n are fixed to the values of quiescent state is the type-I neuronal
model with a strict threshold. If a sufficiently large stimulus is applied to a neuron
in a quiescent state v1 , the state point moves rightwards beyond the stable manifold
of v2 and then goes towards the rightmost equilibrium v3 which corresponds to the
depolarized state of a neuron. If a membrane potential v increases, h is decreased
(after a slight delay) because the variable h tends to the function h1 .v/ which is a
decreasing function of v.
Figure 2.6c is the phase plane with a smaller value of h. In the v–m subsystem
(2.5), the v-nullcline does depend on both h and n while the m-nullcline does not
depend on them. From (2.5a), the v-nullcline is obtained by

gN K n4 .v  VK / C gN L .v  VL /
m3 D (2.6)
gN Na h.VNa  v/

from which we can see that the decrease of h moves the v-nullcline upward
(the shape of v-nullcline is also changed). Note that we consider this equation in the
range of 0  m  1, thus the right-hand side is positive. As a result of displacement
of the v-nullcline, three intersections of the v-, m-nullclines become more clearly
46 2 The Hodgkin–Huxley Theory of Neuronal Excitation

Fig. 2.7 v–m Phase plane of dv dn dh

= = =0
the HH equations (2.3) dt dt dt
1.0

0.8
dm
m
0.6 =0
dt
0.4

0.2

0.0
0 20 40 60 80 100
v (mV)

distinguishable. As is seen from Fig. 2.5b, in the depolarized (high-voltage) region

of v,
n .v/ is slightly larger than
h .v/. Thus n increases subsequently to the de-
crease of h (practically, these changes occur almost simultaneously) since n1 .v/ is
a increasing function of v.
Figure 2.6d shows the case that n is increased in addition to the decrease of h. As
is seen from (2.6), the further increase of n moves the v-nullcline upward. Then two
equilibria v2 and v3 disappear by a saddle-node bifurcation and only equilibrium
v1 remains. As a result of disappearance of those equilibria, the state point near the
equilibrium v3 (depolarized state) cannot stay there and then changes its direction
toward the equilibrium v1 (quiescent state). After this process, h and n change to
increase and decrease, respectively, and then return to the state of Fig. 2.6a.
Figure 2.7 shows the v–m phase plane of the full HH equations (2.3) (projec-
tion of the four-dimensional state space to v–m phase plane). The sigmoid-like
curve is the m-nullcline and is same as the case of v–m subsystem. The
other curve denotes the intersection of v; n; h-nullclines; a curve such that
d v=dt D d n=dt D dh=dt D 0:

gN K fn1 .v/g4 .v  VK / C gN L .v  VL /
m3 D :
gN Na h1 .v/.VNa  v/

The intersection of these two nullclines corresponds to the equilibrium point of the
original HH equations (2.3) and thus the HH equations have a unique equilibrium.
This implies that in a strict sense the original HH equations are a type-II neuronal
model without a distinct threshold. As is seen above, however, in a very short time
range (i.e. if we ignore the temporal change of h and n) the HH equations behave
like a type-I neuronal model which has a distinct threshold. Thus we can understand
why the HH equations have a relatively sharp threshold although they do not have
any threshold in a strict sense.
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 47

2.3.3 Dynamic I–V Relation of the Squid Giant Axon Membrane

In this subsection, let us investigate the HH equations from more global viewpoint
without entering into its detailed dynamics. Figure 2.8 shows dynamic or transient
current–voltage relation of the HH equations; total membrane currents G when a
time t elapses after the membrane voltage is instantaneously changed to v (mV)
from the quiescent state are plotted for various v values. These current–voltage
relations are shown for various values of the time t. At a time t D 0:02 ms, the
relation is almost linear. After some time elapses, negative-resistor characteristics
appear and the current–voltage relation becomes N-shaped. After sufficiently long
time elapses, the relation shows a rectifier characteristics in which only outward
current can be flowed.

G (µA/cm2) t = 0.02 t = 0.05

20 20
0
0
–20
– 20 –40
– 40 –60
– 60 –80
–100
– 80 –120
– 40 0 40 80 120 – 40 0 40 80 120
v (mV)

t = 0.07 t = 1.0

0 1000
500
– 50
0
– 100 – 500
–1000
– 150
–1500
–40 0 40 80 120
– 40 0 40 80 120

t = 4.0 t = 10.0
0 0

– 1000 –1000

– 2000 –2000

– 3000 –3000

– 4000 –4000
–40 0 40 80 120 –40 0 40 80 120

Fig. 2.8 Transient current–voltage relations of the Hodgkin–Huxley membrane

48 2 The Hodgkin–Huxley Theory of Neuronal Excitation

In order to see how these current–voltage relations are generated, let us

decompose the total membrane current to its components: INa and IK where the
(small) leak current is ignored. Figure 2.9 is the similar voltage–current relations to
Fig. 2.8 for Na and K channels. When t D 0:02, the relation of K channel (dotted
curve) is almost linear. After time elapses, the K current in a low-voltage range be-
comes small and finally, the current–voltage relation becomes a rectifier. In the case
of Na channel (solid curve), when t D 0:02, it does not flow much current (almost
flat). After suitable time elapses, it shows large nonlinear characteristics and then
finally becomes almost flat again. The combination of these two current–voltage
relations makes the I –V relation of the total current in Fig. 2.8.
The above current–voltage relations are obtained by numerically solving the
HH equations (2.3). The relation when t D 10:0 is considered as a steady-state
current–voltage relation. This relation can be obtained directly (without numerical
simulation) from the HH equations (2.3) as

G.v; m1 .v/; n1 .v/; h1 .v//:

INa, IK(µA/cm2) t = 0.02 t = 0.05

10 40
0
20
–10 Na
0
–20
–30 –20
K
–40 –40
–50 –60

–40 0 40 80 120 –40 0 40 80 120

v (mV)

t = 1.0 t = 0.07

1000 100
500 50
0
–500 0
–1000 –50
–1500
–40 0 40 80 120 –40 0 40 80 120

t = 4.0 t = 10.0
0 0

–1000 –1000

–2000 –2000

–3000 –3000

–4000 –4000
–40 0 40 80 120 –40 0 40 80 120

Fig. 2.9 Transient current–voltage relations of Na and K channels

2.3 Nonlinear Dynamical Analysis of the Original HH Equations 49

Fig. 2.10 Steady-state G (µA/cm2)

current–voltage relation of 0
the Hodgkin–Huxley
–1000
equations
–2000
–3000
–4000
–5000

–50 0 50 100 150

v (mV)

a b
4000 1.0
3000

G (mA/cm2)
0.5
G (mA/cm2)

2000
1000 0.0
0
–1000 – 0.5
–2000
–1.0
–50 0 50 100 150 –50 –40 –30 –20 –10 0 10 20
v (mV) v (mV)

Fig. 2.11 (a) Steady-state current–voltage relation of the v; m-subsystem of the HH equations.
(b) Magnification of (a)

This steady-state current is plotted as a function of v in Fig. 2.10. Similarly to the

t D 10 case of Fig. 2.8, we can see the rectifier characteristics.
Figure 2.11 shows the steady-state current of the v–m subsystem (2.5):

G.v; m1 .v/; n ; h /

as a function of v where n and h denote the quiescent-state values of n and h,

respectively. Panel (b) is the magnification of (a), from which the current–voltage
curve intersect with the horizontal line at three points which correspond to the three
equilibrium points of the phase plane of Fig. 2.6a. From this observation, we can
also see the threshold property of the HH equations.
Comparing Fig. 2.11 to the transient current–voltage relation of Fig. 2.8, we con-
firm that the v–m subsystem approximates the dynamics of the full HH equations
(2.3) in the time scale of t  1 ms.

2.3.4 Dimension Reduction of the HH Equations

In the v–m subsystem analysis, we investigated the v–m phase plane of the HH
equations by fixing the values of n and h. Namely, we explored the dynamics of
HH equations by decomposing the four-dimensional full phase space into several
n; h-fixed slices.
50 2 The Hodgkin–Huxley Theory of Neuronal Excitation

a b
1.0 1.0
0.8 0.8
0.6 0.6
m n
0.4 0.4
0.2 0.2
0.0 0.0
– 50 0 50 100 150 0.0 0.2 0.4 0.6 0.8 1.0
v (mV) h

Fig. 2.12 Projections of the orbit (solution) of the HH equations to (a) a v–m phase plane and to
(b) a h–n phase plane

This subsection briefly describes the method which reduces the full
four-dimensional system into two-dimensional system (FitzHugh 1961; Krinskii
and Kokoz 1973; Kokoz and Krinskii 1973; Rinzel 1985). Although such dimen-
sion reduction, differently from the v–m subsystem analysis, loses some information
on original dynamics, it is useful to catch the essential feature of the whole four-
dimensional dynamics of the HH equations on a reduced phase plane.
Figure 2.12 shows the projections of an orbit (solution) .v.t/; m.t/; h.t/; n.t//
of the HH equations to (a) a v–m phase plane and to (b) a h–n phase plane. From
panel (a), we can see that the orbit in the region of m  0 or m  1 moves close to
the m-nullcline m D m1 .v/ (broken curve). Panel (b) shows that the orbit moves
restricted in a certain line on the h–n phase plane.
From these observations, we reduce the HH equations (2.3) in two steps:
1. Suppose that the variable m which follows (2.3b) is settled in its steady-state
value: m D m1 .v/ since m is the fast-changing variable (
m is small). Thus we
ignore (2.3b) and substitute m1 .v/ for m in (2.3a).
2. Approximate the orbit on the h–n phase plane by a line n D 0:8.1  h/; we
consider that the variable n linearly depends on h. Thus we ignore (2.3c) and
substitute 0:8.1  h/ for n in (2.3a).
From these reduction steps, we obtain the reduced equations:

dv
C D G.v; m1 .v/; 0:8.1  h/; h/ C Iext ; (2.7a)
dt
dh
D ˛h .v/.1  h/  ˇh .v/h: (2.7b)
dt

This model has only two dynamic variables v and h, and thus has an advantage
that we can analyze the neuronal dynamics on a phase plane rather than the four-
dimensional phase space of the original HH equations.
Figure 2.13a shows an example of the membrane potential waveforms of the orig-
inal HH equations (2.3) (dotted curve) and of the reduced model (2.7) (solid curve).
The upstroke of the membrane potential of the reduced model is slightly faster than
that of the HH equations and the peak value is also bigger than the HH equations.
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 51

a 50
100
80 40

Iext ( mA/cm2)
60 30
v (mV) 40
20 20

0 10
–20
0
–40
0 10 20 30 40 50 60
t (ms)

b
0.7 dh
=0 dv
dt =0
0.6 dt

0.5

0.4
h

0.3

0.2

0.1

0.0
0 20 40 60 80 100
v (mV)

Fig. 2.13 Comparison of (a) membrane potential waveforms and (b) solution orbits in the phase
plane, between the dimension-reduced model (solid curve) and the original HH model (dotted
curve)

This is because we substituted m1 .v/ for m and thus the activation process of Na
channel has been slightly accelerated. Except for these small differences, the two
waveforms of the original and the reduced model are quite similar. This similarity
is surprising since the dimensions of the original and the reduced model are much
different.
Figure 2.13b shows the solution orbit of both models on the v–h phase plane.
The solid curve denotes the orbit of the reduced model (2.7), and the dotted curve
the original HH equations (2.3). The v-nullcline

G.v; m1 .v/; 0:8.1  h/; h/ D gN Na .m1 .v//3 h.VNa  v/

CgN K .0:8.1  h//4 .VK  v/ C gN L .VL  v/ D 0

and the h-nullcline

˛h .v/
h D h1 .v/ D
˛h .v/ C ˇh .v/
are also shown by broken curves. We can see the similarity of orbits of the two
models except for the slight difference in the action potential upstroke.
52 2 The Hodgkin–Huxley Theory of Neuronal Excitation

Topological features (one nullcline is N-shaped and the other is monotonic) of

the phase plane of the reduced model resemble to that of the Bonhoeffer–van der Pol
(BVP) or FitzHugh–Nagumo (FHN) model (FitzHugh 1961; Nagumo et al. 1962).
In fact, the assumptions (i) and (ii) which are used to derive the reduced model (2.7)
(Krinskii and Kokoz 1973; Rinzel 1985) is essentially the same as the one used by
FitzHugh to derive the BVP model (FitzHugh 1961). The reduced model (2.7) by
Krinskii and Kokoz (1973) and by Rinzel (1985) is derived by some logical process
while the BVP model is derived a priori. Thus, the relation of physiological parame-
ters of the original and the simplified models are much clear in the model (2.7) rather
than in the BVP model. For more systematic reduction of general HH-type models,
see Abbott and Kepler (1990), Golomb et al. (1993), and Kepler et al. (1992).

2.3.5 Bifurcation Analysis

In this subsection, we consider the dependence of the HH equations’ behavior on the

parameter Iext ; the constant-current-transfer characteristic of the HH neuron (Rinzel
1978; Guttman et al. 1980). The neuron model produces an action potential with
response to an external pulsatile stimulus. We can expect that the neuron generates
action potentials persistently when a continuous current is applied externally.
Figure 2.14 shows the example of such repetitive firings (oscillation or periodic
orbit) of the HH neuron when Iext D 7. At t D 62, a pulsatile stimulus is applied to
the neuron in addition to the constant current injection. After the pulse stimulus, the
repetitive firing is stopped in spite of the persistent injection of the constant current.
This means that an repetitive firing (stable limit cycle) and a quiescent state (stable
equilibrium) coexist when Iext D 7. The state point of the HH neuron is moved from
one attractor (limit cycle) to an another one (equilibrium) by the external pulse.
We note that the applied pulse is depolarizing current; not only inhibitory input
but also excitatory input can inhibit such repetitive firing since it is a nonlinear
oscillation. The timing (phase) when the pulse is applied is important.

100 20

80 18
60 16
Iext ( mA/cm2)

40 v
v (mV)

14
20 12
0 10
Iext
–20
8
–40
6
–60
20 40 60 80 100
t (ms)

Fig. 2.14 Example of a membrane potential waveform of the HH equations when an external
constant current is applied: Iext D 7 A cm2 . A pulse is applied to the model in addition to the
constant current at t D 62
2.3 Nonlinear Dynamical Analysis of the Original HH Equations 53

a b
v (mV)
100. 125.

DC3 100. DC3

75.
75.
50.
50.
DC2
25. DC1 DC2
25. DC1
HB1 HB1
HB2
0. 0.
0. 50. 100. 150. 200. 0.0 5.0 10.0 15.0 20.0
25. 75. 125. 175. 2.5 7.5 12.5 17.5
2 2
Iext (µA/cm ) Iext (µA/cm )
c v (mV)
25.0
22.5
A PD2
20.0
DC2
17.5
15.0 PD1
DC1
PD3
12.5
10.0
7.80 7.85 7.90 7.95 8.00 8.05
Iext (µA/cm2)
d e
15.0 25.
12.5 DC1 20.
10.0 15.
7.5
10.
5.0
5. A
2.5
0.0 0.
–2.5 –5.
–5.0 –10.
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
0.10 0.30 0.50 0.70 0.90 0.10 0.30 0.50 0.70 0.90
t (ms) t (ms)

Fig. 2.15 (a) One-parameter bifurcation diagram of the HH equations. (b) Magnification of (a).
(c) Magnification of (b). (d, e) Membrane potential waveforms at the points DC1 and A of (c)

Figure 2.15a shows the dependence of the solution of the HH equations on the
parameter Iext ; the v values of the stationary solution of the HH equations are plot-
ted for various values of Iext where the maximum value of v is plotted for a periodic
(oscillatory) solution. Solid and dotted curves denote stable and unstable equilib-
ria, respectively. The filled (open) circles denote stable (unstable, resp.) periodic
solutions.
Panel (b) is the magnification of left part of (a) and we can verify the multi-
stability of an equilibrium and a periodic solution when Iext D 7 (ref. Fig. 2.14). At
the point HB2 of panel (a), a stable periodic solution bifurcates from a equilibrium
point by the (super-critical or stable) Hopf bifurcation. An unstable periodic solution
54 2 The Hodgkin–Huxley Theory of Neuronal Excitation

is bifurcated by the (sub-critical or unstable) Hopf bifurcation at the point HB1. As

is seen from panel (b), a multistability occurs near the sub-critical Hopf bifurcation.
At the point DC3, a double-cycle bifurcation or saddle-node bifurcation of periodics
occurs and a pair of stable and unstable periodic solutions is generated.
At both points DC1 and DC2, double-cycle bifurcations occur also. Near Iext D
7:9, four periodic solutions (one stable solution and three unstable solutions) coexist
(Rinzel and Miller 1980). The fact is that there are more coexisting (unstable) so-
lutions. Panel (c) is the magnification of the region near the points DC1 and DC2 of
(b). (Note that the periodic solutions are denoted by curves rather than circles.) At
the points PD1 and PD2, period-doubling bifurcations of unstable periodic solution
occur. Panel (d) shows the membrane potential waveform (only one-period length
of periodic solution is shown and abscissa is the time normalized by its period) of
such unstable periodic solution at the point DC1 of panel (c). Panel (e) is the similar
waveform of the period-doubled solution at the point A of panel (c).
We can not observe such unstable solutions in real experiments. The unstable
solutions, however, connect “the missing links” between stable solutions and much
help us to understand the total behavior of neuronal models. The bifurcation analysis
of Fig. 2.15 is made by the use of numerical bifurcation analysis software called
AUTO (Doedel et al. 1995). AUTO is very useful software which can detect several
bifurcation points automatically and can trace both stable and unstable branches of
equilibria and periodic solutions.
Figure 2.16a shows the period of the periodic solutions shown in Fig. 2.15a.
Panel (b) is the magnification of panel (a). The period of stable periodic solutions
(closed circle) varies in the range from several milliseconds to 20 ms. The period
does not change much totally although the variation is comparatively large in the
small Iext range.

a (ms)
b (ms)
30. 30.

25. 25.
DC3 DC2
20. 20. DC1
DC2
15. DC1 15. DC3

10. 10.

5. 5. HB1
HB1 HB2
0. 0.
0. 25. 50. 75. 100. 125. 150. 175. 6.0 7.0 8.0 9.0 10.0
6.5 7.5 8.5 9.5
Iext(µA/cm2) Iext(µA/cm2)
.

Fig. 2.16 (a) Period of the periodic solution shown in Fig. 2.15a and (b) Magnification of (a)
http://www.springer.com/978-4-431-53861-5